This version strays from the first version in a couple ways:
- Instead of a bunch of 0/1 pixel readings, the state is instead 3 distances, which is a step towards a real-world scenario where we'll use sonar to detect the state of the room.
- Obstacles are no longer randomly created, but they now move in an effort to combat over-fitting.
This is a bit of a "step" version, as in the next version, we'll be focused entirely on trying to apply this simulation to a real RC car.